Tavily Search API
AI-optimized web search API designed for LLM agents that returns clean, pre-extracted textual content from search results rather than raw HTML or SERP metadata.
Score Breakdown
⚙ Agent Friendliness
🔒 Security
Standard API key model. No PII or user data stored beyond search queries for rate limiting. MCP server security depends on transport configuration at deployment.
⚡ Reliability
Best When
You need web search results that are immediately consumable by an LLM without HTML parsing, chunking, or cleaning steps.
Avoid When
Your agent needs raw SERP data, full HTML, or search metadata beyond extracted text and URLs.
Use Cases
- • Ground LLM agent responses with real-time web information by injecting Tavily search results directly into context
- • Build RAG pipelines that retrieve fresh web content as an alternative to stale vector store documents
- • Perform automated research tasks where an agent collects factual information from multiple web sources in a single call
- • Validate or fact-check LLM-generated claims against current web content retrieved via structured search
- • Enable agents to monitor topics by running periodic searches and comparing result sets for changes
Not For
- • Applications needing full raw HTML or complete SERP metadata (ad positions, pagination data, knowledge panels)
- • High-volume anonymous search at millions of queries per day where cost-per-query is the dominant constraint
- • Use cases requiring search results with guaranteed original source formatting preserved
Interface
Authentication
API key passed as Bearer token in Authorization header or as api_key field in JSON body. No scope granularity; single key grants full account access.
Pricing
Credits consumed per search call regardless of result count. The 'extract' endpoint for deep content extraction consumes more credits per call than basic search.
Agent Metadata
Known Gotchas
- ⚠ The max_results parameter caps results per call but does not guarantee that many results will be returned; agents should not assume result count equals max_results
- ⚠ Content extraction quality varies significantly by target site; paywalled, JavaScript-heavy, or bot-protected pages may return empty or partial content without an explicit error
- ⚠ The 'search_depth' parameter (basic vs advanced) significantly affects both quality and credit consumption; advanced mode uses more credits per call and should be reserved for tasks where depth matters
- ⚠ MCP server tool names differ slightly from REST API parameter names; agents switching between MCP and REST integrations must reconcile the schema differences
- ⚠ Results include a score field per result but the scoring methodology is not documented; agents should not use this score as an authoritative relevance metric without validation
Alternatives
Full Evaluation Report
Detailed scoring breakdown, competitive positioning, security analysis, and improvement recommendations for Tavily Search API.
Scores are editorial opinions as of 2026-03-06.